80 On a Possible Explanation of the Method emj^loyed hy Ndbert 
In this way it is easy to measure not only the maximum value, 
but the values corresponding to every division of the screw-head. 
If an objective of high power is employed, care is necessary to have 
the surfaces of both pieces of glass as nearly plane as possible. 
Much better results are obtained by using a piece of cover-glass not 
larger than tV of a square inch. 
Mr. John M. Blake, of New Haven, did me the kindness to 
photograph on cover-glass, the plate whose measures are given in 
column III. By reversing the plates, in the way indicated above, 
he found almost precisely the same value for the maximum periodic 
error deduced above ; viz. -^-^ of an inch. 
In passing, it may be interesting to note that though the lines 
of the Kutherford plate are more distinct than those of the Nobert 
plate, and though the errors of spacing are considerably less, yet 
the former was rejected from the start as an imperfect one, while the 
latter gives excellent results, yet both plates will show with about 
equal distinctness four lines between the components of the magne- 
sium line h. This is hardly in accordance with the theory that 
the optical test of parallelism of lines, and of equality of spacing, is 
far more perfect than the test of actual measurement. It is evident 
that the theoretical limit of accuracy required, in order to produce 
the solar lines in the greatest perfection, has rarely if ever been 
reached in actual practice. All the evidence seems to point to the 
conclusion that the brilliancy of the spectrum depends as much on 
the character of the lines, and especially on the character of the 
edges, as on the equality of the spacing. 
It is obvious, then, that the errors which are to be the most 
feared, both on account of their magnitude and the likelihood of 
their escaping detection, are those which are periodic in their 
character. To the investigation of the sources of these errors, in 
my own machine, several months of careful study have been given. 
Without entering into a detailed account of fruitless experiments, I 
will give only the conclusion at which I have arrived, viz. that the 
jperiodicity resides, not in the screw itself, hut in the mounting of 
the serew. The evidence on this point seems to be conclusive. In 
a large number of separate measurements extending over several 
weeks, substantially the same system of values as those given in 
column III. were found. These values were also constant for 
different parts of the screw. Conjecturing that the trouble might 
arise from unequal friction between the screw and the nut at different 
parts of the revolution, owing to the want of parallelism between 
the screw and the fixed way on the bed of the machine a slight 
movement was given to tbe adjusting screws, which clamped the 
split nut. At once the system of corrections was wholly changed, 
not only in value but in sign, and the values now found, remained 
constant under every variety of tests applied. After a few weeks, 
